In this paper, we present a new approach that uses a combination of a compliant robot foot with a flexible tactile-array sensor to classify different types of cylindrical terrains. The foot and sensor were installed on a robot leg. Due to their compliance and flexibility, they can passively adapt their shape to the terrains and simultaneously provide pressure feedback during walking. We applied two different methods, which are average and maximum value methods, to classify the terrains based on the feedback information. To test the approach, We performed two experimental conditions which are (1) different diameters and different materials and (2) different materials with the same cylindrical diameter. In total, we use here eleven cylindrical terrains with different diameters and materials (i.e., a 8.2-cm diameter PVC cylinder, a 7.5-cm diameter PVC cylinder, a 5.5-cm diameter PVC cylinder, a 4.4-cm diameter PVC cylinder, a 7.5-cm diameter hard paper cylinder, a 7.4-cm diameter hard paper cylinder, a 5.5-cm diameter hard paper cylinder, a 20-cm diameter sponge cylinder, a 15-cm diameter sponge cylinder, a 7.5-cm diameter sponge cylinder, and a 5.5-cm diameter sponge cylinder). The experimental results show that we can successfully classify all terrains for the maximum value method. This approach can be applied to allow a legged robot to not only walk on cylindrical terrains but also recognize the terrain feature. It thereby extends the operational range the robot towards cylinder/pipeline inspection.